Well, pg's solution works for the limited input provided and you haven't given any further particulars regarding input. That solution breaks just changing the first word from "thought" to "though" :

my $var = "though test tot 1 2 3 tesset";
$var =~ s/(t.*?t)/($1 ne "test") ? "" : $1/ge;
print $var;  # prints: esoesset
[download]

But, now you mention a further constraint that the words to be deleted may not contain any 't's inside, which is not inferrable from your earlier posts at all. Providing a good specification is much more than providing a sample case (but providing test cases *is* important).

Anyway, here's a go at your new specs:

my $var = <<TT;
target blah foo test thought 123 though tempest
testament though tightest treatment thermostat tantamount taboo
TT
$var =~ s/(?!\btest\b)(\bt[^t\W]*t\b)//g;
print $var;
__END__
## Result:
 blah foo test  123 though 
testament though tightest treatment thermostat tantamount taboo
[download]

So, all the 't.*t' words on the second line remain because they contain a 't' character within. All the 't.*t' words on the first line get deleted except for 'test'.

OK so that's \bt[^t]+t\b -- word-boundary, then a t, then one or more other characters not a t, then a t, then a word boundary.

Apart from the abbreviation "tt" this should be fine.

So "tent", "tesseract", "tot", "tort" and "test" itself will match this pattern.

However, "testament" will fail it because of the "t" in the middle.

Then you need a special case for "test" itself, which you can do with the /e modifier and the ternary operator, as in pg's example above.

So something like this:

#!/usr/bin/perl -w
use strict;

my $words='test Buffy testament Anya tot Willow tesseract
 Faith tent';
$words =~ s/\b(t[^t]+t)\b/$1 eq "test" ? $1  : ''/ge; 
print $words;
# prints 'test Buffy testament Anya  Willow  Faith';
[download]

Where the regex means "Find words matching t, something-not-t, then t at the end. Replace them with nothing, unless they're the word test, in which case, replace them with themselves".

You could replace the ternary thing with this more longwinded version if you liked:

$words =~ s/\b(t[^t]+t)\b/
           my $temp = $1;
           if($temp eq 'test'){
             $temp
           }else{
             ''
           }/xge;
[download]

Your character class of [^t] can itself cross word boundaries so that strings like: "this will be a problem right?" will be a problem, right?

Quite right, very true, didn't think of that.

What if we add a not-whitespace-either to the character class?

#!/usr/bin/perl -w
use strict;

my $words='test Buffy testament Anya tot Willow tesseract
 Faith tent this is a problem right?';
$words =~ s/\b(t[^t\s]+t)\b/$1 eq "test" ? $1  : ''/ge; 
print $words;
[download]

---




($_='kkvvttuubbooppuuiiffssqqffssmmiibbddllffss')
 =~y~b-v~a-z~s;                             print
[download]

I spent quite a bit of time trying to craft my example carefully, so that if there was a regex solution to return the result I specified, I'd have my answer.

But the problem is that you left it at the example. I could have given you a couple of regexes that solved your example, but would probably have failed to do what you wanted on the second example you tried.

pg got it perfectly.

I want to delete any words that start with "t", end with "t", but do not contain any other "t"s within, except for the word "test".

s/(t.*?t)/($1 ne "test") ? "" : $1/ge;
[download]

    $_ = "this is the wristwatch";
    s/(t.*?t)/($1 ne "test") ? "" : $1/ge;
    print;
    __END__
    he wrisch
[download]

Abigail

s/\s*\bt(?!est)[^t\W]*t\b//g;
[download]